AMD GPU support for conv, gemm, rnn for fp32.

[dagamayank]

This PR adds the support for AMD GPUs for convolutions, gemm, and RNN benchmarks.

• This is dependent on ROCm v.1.7.1 and MIOpen v1.3
• RNN benchmarks will only work on AMD GPUs with 16GB memory.

[sharannarang]

Thanks for creating this PR! We're excited to have support for AMD in DeepBench.

I'll review the code later this week. A couple of questions/thoughts:

1. Are you planning to add results for AMD GPUs as well?
2. Could you also update the README with instructions on how to build the code? Something similar to Nvidia Benchmarks would be great.

[dagamayank]

@sharannarang

A couple of questions/thoughts:

• Are you planning to add results for AMD GPUs as well?
• Could you also update the README with instructions on how to build the code? Something similar to Nvidia Benchmarks would be great.

Done.

[dagamayank]

@sharannarang Ping.